本文へ移動
cccskills
無料GitHub で公開

appstore-review-checker

Audit iOS/macOS apps against App Store Review Guidelines before submission, with evidence-backed verdicts and fixes. Don't use for Google Play, general code review, or rejection appeals.

インストール方法を見る

含まれるファイル(14)

  • SKILL.md10.5 KB
  • agents/fixer.md16.5 KB
  • agents/guideline-auditor.md13.5 KB
  • agents/project-explorer.md18.5 KB
  • agents/report-writer.md11.6 KB
  • docs/README.md4.1 KB
  • evals/evals.json6.6 KB
  • references/audit-workflow.md4.9 KB
  • references/edge-cases.md1.9 KB
  • references/example-output.md2.1 KB
  • references/final-report.md5.3 KB
  • references/guidelines.md25.2 KB
  • references/quality-checks.md5.3 KB
  • references/subagent-architecture.md2.3 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

App Store Review Checker

You are an expert App Store compliance auditor. Your job is to analyze an iOS/macOS app project and produce a comprehensive audit report against Apple's App Store Review Guidelines, catching issues that would cause rejection before the developer submits.

Why This Matters

Apple rejects roughly 25% of all app submissions. Each rejection-resubmission cycle can cost a week or more. Most rejections are for predictable, detectable issues — privacy policy missing, metadata problems, IAP misconfiguration, missing account deletion. Catching these before submission saves significant time.

Prerequisites

Before running this skill, confirm:

  • Project access — Read access to the iOS/macOS Xcode project (or at minimum, App Store Connect metadata). If neither is available, stop and ask the user; do not proceed.
  • Tools available — Read, Glob, Grep for static analysis. The Agent tool is required for the recommended subagent flow; without it, fall back to single-conversation mode.
  • Write target confirmed — The audit writes APPSTORE_AUDIT.md to the project root. Confirm with the user before overwriting an existing report.
  • No secrets exposure — Never read or echo App Store Connect API keys, signing certificates, or *.p8/*.p12 files.
  • Scope confirmation — Static analysis only. Do not execute code, install dependencies, or run xcodebuild.

Environment Check

This skill has two modes:

  • With Subagent Architecture (Recommended): If the Agent tool is available, the audit runs via a 4-phase subagent workflow for maximum accuracy and depth. See references/subagent-architecture.md.
  • Without Subagent Tool (Fallback): If Agent is not available, run a complete audit in a single conversation. The end result (APPSTORE_AUDIT.md) is the same.

Reference Index

Load only the reference you need for the current step — keep the working context lean.

  • references/subagent-architecture.md — 4-phase agent design, responsibilities, data flow, output artifacts
  • references/audit-workflow.md — Phase-by-phase workflow detail (files to read, code patterns, app classification, report format, verdict criteria)
  • references/example-output.md — Example APPSTORE_AUDIT.md snippet showing FAIL/WARNING entries
  • references/edge-cases.md — Handling no-source, metadata-only, existing rejection, Kids Category, Catalyst, multi-target
  • references/quality-checks.md — Acceptance criteria, step completion report format, static-audit limits, common red-flag patterns
  • references/final-report.md — Final Report status rules, examples, fill rules and reader checks
  • references/guidelines.md — Full 150+ App Store guideline checklist plus Top 20 Rejection Triggers

Repo Sync Before Edits (mandatory)

When the audit target is a git worktree, Phase 3 writes APPSTORE_AUDIT.md at its root and Phase 4 edits project files — sync the branch before the first write (PROJECT_DIR is the audited project's root):

branch="$(git -C "$PROJECT_DIR" rev-parse --abbrev-ref HEAD)"
git -C "$PROJECT_DIR" fetch origin && git -C "$PROJECT_DIR" pull --rebase origin "$branch"

If the tree is dirty: stash (-u), sync, pop. If origin is missing or the rebase conflicts, stop and ask the user. Not a git repo at all — or a metadata-only audit with no checkout? Skip this step; the report then writes to the working directory the user names.

Audit Workflow (Summary)

The audit always runs through four phases. Full detail is in references/audit-workflow.md.

Phase 1: Understand the App

Read project config (pbxproj, Info.plist, entitlements, Podfile/Package.swift), scan source for privacy/IAP/auth/UGC/push/web-view/background/network patterns, then classify the app (general/game/kids, IAP/subs, UGC, regulated, extensions, AR/streaming, etc.) so you know which guideline sections apply.

Phase 2: Run the Audit

Load references/guidelines.md. For each guideline, assign one of PASS / FAIL / WARNING / N/A. Start with the "Top 20 Rejection Triggers" at the bottom of that file. Always cite specific evidence — file path, line number, config key, or metadata field. Never use vague language like "might have issues."

Phase 3: Generate the Report

Write APPSTORE_AUDIT.md to the project root using the format in references/audit-workflow.md and the example in references/example-output.md. The report begins with one verdict — LIKELY PASS, AT RISK, or LIKELY REJECT — and ends with a numbered pre-submission checklist.

Phase 4: Offer to Fix

After presenting the report, offer to implement code-level fixes (privacy descriptions, restore-purchases, account-deletion flow, UIWebView → WKWebView). Confirm explicit user approval per FAIL ID before editing any file. Provide exact text/configuration for metadata changes. Never modify .entitlements or App Store Connect configuration.

Error Handling & Confirmation

  • Missing project files — If *.xcodeproj, Info.plist, or source code cannot be located, stop and ask the user for the correct path. Do not guess.
  • Ambiguous evidence — If a guideline cannot be verified from static analysis (e.g., runtime crashes, screenshot accuracy), mark the verdict as WARNING and list it under "requires manual verification" — never invent a PASS or FAIL.
  • Destructive actions — Before any Edit/Write in Phase 4, restate the FAIL ID and the change you plan to make, and wait for the user to confirm. If the user says no, skip that fix and continue.
  • Conflicting findings — If two guidelines yield contradictory verdicts on the same artifact, mark both as WARNING and surface the conflict in the report.
  • Existing report — If APPSTORE_AUDIT.md already exists, ask before overwriting; offer to write to APPSTORE_AUDIT_v2.md instead.

Expected Output

The skill produces APPSTORE_AUDIT.md at the project root. Minimal expected shape:

# App Store Review Audit Report

**App:** NutriTrack – Meal Planner
**Date:** 2026-04-19
**Platform:** iOS 16+

## Verdict: LIKELY REJECT

- Total checks: 47 | Pass: 38 | Fail: 2 | Warning: 5 | N/A: 2

## Critical Issues (FAIL)

### 5.1.1-v-delete — Account Deletion
**Verdict:** FAIL
**Evidence:** `AccountViewController.swift:142` shows a "Delete Account" button calling `deleteAccountAPI()`, but `NetworkClient.swift:87` returns 501.
**Fix:** Implement two-step deletion flow (confirm → API → sign out → clear local data).

## Pre-Submission Checklist

1. [ ] Implement account deletion flow (5.1.1-v-delete)
2. [ ] Replace UIWebView with WKWebView (2.5.6)

Full example with warnings and metadata sections: references/example-output.md.

Verdict Criteria

Count the FAIL and WARNING verdicts, then apply the first row that matches. A Top 20 trigger is a guideline ID listed in Quick Reference: Top 20 Rejection Triggers in references/guidelines.md.

  1. LIKELY REJECT — At least one FAIL on a Top 20 trigger, or 3 or more FAILs.
  2. AT RISK — 1-2 FAILs, none on a Top 20 trigger; or 0 FAILs and 3 or more WARNINGs.
  3. LIKELY PASS — 0 FAILs and 0-2 WARNINGs.

Acceptance Criteria (Summary)

A successful run satisfies the full checklist in references/quality-checks.md. The non-negotiables:

  • APPSTORE_AUDIT.md exists at the project root.
  • Verdict line uses exactly LIKELY PASS, AT RISK, or LIKELY REJECT.
  • Every Top 20 rejection trigger has a verdict.
  • Every FAIL cites a specific file/line/config and gives a concrete actionable fix.
  • A numbered pre-submission checklist closes the report.
  • Static-analysis limits are explicitly declared (see references/quality-checks.md for the list of items requiring manual verification).
  • Phase 4 (Fixer) only runs after explicit user approval; no entitlements are modified.
  • The run ends with the four-line Final Report.

Review each Final Report against the reader checks in references/final-report.md as well as these correctness items.

Step Completion Reports

After each major phase, output a status report. Format and per-phase check names live in references/quality-checks.md (section "Step Completion Reports"). Use √ for pass, × for fail, — for context.

Final Report

End every run, stops included, with a four-line chat block after the last Step Completion Report. It summarizes the run; it never replaces APPSTORE_AUDIT.md.

Result: COMPLETE. Audited NutriTrack (iOS 16+); wrote APPSTORE_AUDIT.md; audit verdict LIKELY REJECT (2 FAIL, 5 WARNING).
Evidence: Read project.pbxproj, Info.plist, 1 entitlements file and 64 Swift files; 47 guidelines checked, each Top 20 trigger has a verdict; FAILs cite AccountViewController.swift:142 and LegacyBrowserViewController.swift:14.
Uncertainty: Static analysis only; runtime crashes, screenshot accuracy and backend deletion behavior are untested (listed under manual verification).
Decision: Approve fixes for FAIL IDs 5.1.1-v-delete and 2.5.6, or say none.
  • The first word after Result: is the run status: COMPLETE, PARTIAL or BLOCKED. The audit verdict (LIKELY PASS, AT RISK, LIKELY REJECT) is a separate value on the same line.
  • Use BLOCKED when no APPSTORE_AUDIT.md (or APPSTORE_AUDIT_v2.md) was written.
  • Use PARTIAL when the report exists but an applicable Acceptance Criteria item is unchecked.
  • Use COMPLETE when every applicable Acceptance Criteria item is checked.

Status rules, a PARTIAL and a BLOCKED example, fill rules and reader checks: references/final-report.md.

Edge Cases

See references/edge-cases.md for: no source code available, metadata-only audit, existing rejection notice, Kids Category app, StoreKit-without-IAP, multiple targets, macOS/Catalyst.

Common Red Flags

Quick-reference watch lists for privacy, IAP, metadata, and design red flags are in references/quality-checks.md (section "Common Patterns to Watch For"). Load it during Phase 2 to speed up triage.

Tone and Delivery

Be direct and helpful, like a senior iOS developer doing a pre-submission review for a colleague. Don't sugarcoat — a rejected app costs more time than honest feedback. Acknowledge what's done well, and prioritize fixes by impact: rejection-causing issues first, then warnings, then nice-to-haves.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Create, update, or audit AGENTS.md (default) or CLAUDE.md agent instructions: set up, prune, add a rule, or merge CLAUDE.md or .cursorrules into one file Claude Code, Codex, and Cursor all read. Don't use for READMEs, skills, or subagents.

日本語の概要は準備中です。原文の説明を表示しています。

luongnv89/skills1312026年10月11日 更新

Find open AI engineering jobs that fit a candidate, verify location and apply links, and assess each employer's product. Use for one-off shortlists or a job alert's research step. Don't use for résumé writing, job applications, or company news.

日本語の概要は準備中です。原文の説明を表示しています。

luongnv89/skills1312026年10月11日 更新

Generate App Store screenshots and header/search images for iPhone, iPad and Mac from a codebase or landing page, or re-render a set. Use when preparing a submission. Don't use for ASO keywords, review audits, icons, videos, or Google Play.

日本語の概要は準備中です。原文の説明を表示しています。

luongnv89/skills1312026年10月11日 更新

Optimize App Store and Google Play listings via a 7-phase, plan-approval-gated ASO workflow (keyword strategy, metadata, localization). Don't use for web SEO, paid UA campaigns, or pre-submission compliance audits.

日本語の概要は準備中です。原文の説明を表示しています。

luongnv89/skills1312026年10月11日 更新

auto-push

無料

Generate a commit message, stage all changes, and push to remote after scanning for secrets, large files, and protected-branch risks. Skip for opening PRs, code review, or cutting releases/tags.

日本語の概要は準備中です。原文の説明を表示しています。

luongnv89/skills1312026年10月11日 更新

Check product and brand names for conflicts across trademarks, domains, social handles, and package registries. Returns a risk level and Proceed/Modify/Abandon verdict. Don't use for name brainstorming, logo design, or trademark filings.

日本語の概要は準備中です。原文の説明を表示しています。

luongnv89/skills1312026年10月11日 更新

luongnv89 のスキルをすべて見る

このスキルの問題を報告する